HAL Id: halshs-01612901
https://halshs.archives-ouvertes.fr/halshs-01612901
Submitted on 17 Apr 2018HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Climate Damage on Production or on Growth: What
Impact on the Social Cost of Carbon?
Céline Guivarch, Antonin Pottier
To cite this version:
Céline Guivarch, Antonin Pottier. Climate Damage on Production or on Growth: What Impact on the Social Cost of Carbon?. Environmental Modeling & Assessment, Springer, 2017, �10.1007/s10666-017-9572-4�. �halshs-01612901�
Climate damage on production or on growth: what impact
on the social cost of carbon?
Céline Guivarch CIRED
Ecole des Ponts ParisTech 45 bis, avenue de la Belle Gabrielle 94736 Nogent-sur-Marne Cedex guivarch@centre-cired.fr Antonin Pottier MINES ParisTech PSL Research University CERNA - Centre for industrial economics i3, CNRS 60, boulevard Saint-Michel 75272 Paris Cedex 06 antonin.pottier@mines-paristech.fr
Note: this text is a post-print from the article: Guivarch, C. and A. Pottier. 2018. Cli-mate Damage on Production or on Growth: What Impact on the Social Cost of Carbon?. Environmental Modeling Assessment 23 (2) : 117-130.
Abstract
Recent articles have investigated with Integrated Assessment Models the possibility that climate damage bears on productivity (TFP) growth and not on production. Here, we compare the impact of these alternative representations of damage on the social cost of carbon (SCC). We ask whether damage on TFP growth leads to higher SCC than damage on production ceteris paribus. To make possible a controlled comparison, we introduce a measure of aggregate damage, or damage strength, based on welfare variations.
With a simple climate-economy model, we compare three damage structures: quadratic damage on production, linear damage on growth and quadratic damage on growth. We show that when damage strength is the same, the ranking of SCC be-tween a model with damage on production and a model with damage on TFP growth is not unequivocal. It depends on welfare parameters such as the utility discount rate or the elasticity of marginal social utility of consumption.
JEL codes: Q51, Q54
Climate damage on production or on growth: what impact
on the social cost of carbon?
April 17, 2018
Abstract
Recent articles have investigated with Integrated Assessment Models the possibility that climate damage bear on productivity (TFP) growth and not on production. Here, we compare the impact of these alternative representations of damage on the social cost of carbon (SCC). We ask whether damage on TFP growth leads to higher SCC than damage on production ceteris paribus. To make possible a controlled comparison, we introduce a measure of aggregate damage, or damage strength, based on welfare variations.
With a simple climate-economy model, we compare three damage structures: quadratic damage on production, linear damage on growth and quadratic damage on growth. We show that when damage strength is the same, the ranking of SCC be-tween a model with damage on production and a model with damage on TFP growth is not unequivocal. It depends on welfare parameters such as the utility discount rate or the elasticity of marginal social utility of consumption.
JEL codes: Q51, Q54
keywords: climate change; damage; social cost of carbon; growth; TFP; IAM
1 Introduction
The social cost of carbon1 (SCC) the present social value of damage from an additional
ton of CO2 released in the atmosphere is an important concept in environmental policy.
It represents the price that should be put on greenhouse gases emissions to maximize welfare in a rst-best world. Indeed, along an optimal abatement path, abatement is such that marginal abatement cost equals the SCC. The SCC informs thus both on the level of expected climate change damage costs and on the level of eort in mitigation policies.
The SCC recently received close scrutiny for its direct role in policy evaluation (Inter-agency Working Group on Social Cost of Carbon (2010), Kopp et al. (2012), Greenstone et al. (2013), Pizer et al. (2014), Committee on Assessing Approaches to Updating the Social Cost of Carbon et al. (2017)). By monetizing the marginal damage associated with an incremental increase in emissions (equivalently the bene ts associated with an incre-mental reduction), the SCC makes it possible to account for the social costs or bene ts of regulatory actions that aect CO2 emissions. It can thus allow to compare policies and
identify those that have positive net bene ts. The values of the SCC currently in use for the evaluation of US federal rulemaking are for the year 2015: 11, 36, 56 and 105 in 2007 dollars per metric ton of CO2(Interagency Working Group on Social Cost of Carbon,
1More accurately the social cost of CO
2 as it relates to the impact of a ton of CO2 and not a ton of
carbon. This is simply a question of changing units (from $ /tCO2 to $ /tC) and not a dierent concept.
2015), the rst three are tied to discount rates of 5, 3, and 2.5 percents, respectively, where the fourth is intended to represent the upper tail of the SCC distribution.
The tools used to quantify the social cost of carbon are called Integrated assessment models (IAMs). IAMs2 balance costs and bene ts of climate change mitigation, in an
eco-nomic model where both the emissions abatement costs and the climate change damage costs are introduced. Following the seminal work of Nordhaus (1992, 1994), damage costs from climate change have been represented in these models with a damage function that reduces the current economic production. In most models, damage thus bears on produc-tion. It is the case in particular in the three models used in the Interagency Working Group on Social Cost of Carbon (2015): DICE (Nordhaus, 2010), FUND (Antho and Tol, 2013) and PAGE (Hope, 2013).
The assumptions underlying this speci cation are questionable. There is a growing amount of evidence that damage may also bear on productivity growth rates rather than on production levels. Indeed some eects of climate change will be permanent, e.g. deaths from extreme events or destruction of ecosystems and land from sea level rise. Econometric studies also give empirical support for damage on GDP growth, e.g. Dell et al. (2009), Dell et al. (2012), Deryugina and Hsiang (2014), Colacito et al. (2014), Hsiang and Jina (2014). For example, Dell et al. (2012) used panel data to assess the impact of climate change on growth. They regressed growth rate on lagged temperatures. They found that temperature eect is signi cant for poor countries, and is persistent over a decade, suggesting an impact on growth rate rather than on production level. Or, Deryugina and Hsiang (2014) estimate the eect of daily temperature on annual income in United States counties and nd that productivity of individual days declines by roughly 1.7% for each 1◦C increase in daily
average temperature above 15◦C. These empirical research suggests that Total Factor
Productivity (TFP) growth can be impacted by climate change and not only production levels. A theoretical argument raised by Moyer et al. (2014) goes in the same direction: in an endogenous growth model, damage on production level translates into damage on TFP. This gives su cient reasons to investigate the impacts on SCC of damage on TFP growth. Pindyck (2011, 2012) studied the eect of damage on growth for a measure of willingness to pay (WTP) to limit climate change to a given temperature increase, and found that WTP is lower than in the case of damage on production. But he did not investigate the eect on the SCC. It is only very recently that, following the recommendation of Revesz et al. (2014) to include damage on growth in IAMs, a few studies have investigated the possibility that climate damage bears on TFP growth, and not (or in addition to) on production (Dietz and Stern, 2015; Moyer et al., 2014; Moore and Diaz, 2015).
Moore and Diaz (2015) consider a two-region economy with damage on TFP growth and production calibrated on empirical data. They found that adding damage on TFP growth multiplies the SCC by a factor 6.7 in their central assumptions, compared to the case with damage on production only. Dietz and Stern (2015) or Moyer et al. (2014) allocate a share of damage costs on TFP growth3. With a share set at 5%, which is
quali ed as conservative by the authors, the SCC is multiplied by a factor 2.7 (Dietz and Stern, 2015), compared to the case with no damage on TFP growth. These studies
2There are two dierent types of IAMs. First, the Cost-Bene t type concerned in this article. Second,
a dierent type of IAMs, sometimes called process-based IAMs, that explicitly represent the drivers and processes of change in global energy and land use systems linked to the broader economy. The IAMs of this second type do not represent damage from climate change, and are used to analyze transformation pathways to achieve a pre-determined level of mitigation eort such as 2◦
C climate stabilization, in a cost-eectiveness framework. Examples of process-based IAMs include the models used to quantify the Shared Socioeconomic Pathways (Riahi et al., 2017).
concluded that damage on TFP growth leads to higher SCC than damage on production. In this article, we ask a dierent question: which damage structure (on TFP growth or on production) leads to a higher SCC? Our approach is similar to Pottier et al. (2015), who investigate how the form of the damage function (quadratic vs. sigmoid) impacts the optimal SCC and abatement trajectories. To compare alternative structures of climate damage, we need to disentangle the eect of the structure (i.e. the functional form of the damage function and whether it bears on production or on growth) and the eect of the magnitude of damage.
Indeed, increasing the magnitude of damage increases the SCC, even without changing the damage structure. For example, in a model with damage on production: if we increase some parameter in the damage function, the SCC will increase too, and would eventually become larger than any SCC computed with a model with damage on growth with given parameters. So the SCC depends not only on the damage structure, but also on the magnitude of damage. The two are conceptually distinct and should not be confused. To put it dierently, increasing some parameter of the damage function generally increases both the marginal damage, which the SCC is a measure of, and aggregate damage. To investigate the eect of a damage structure, we therefore have to track both the marginal damage (the SCC) and aggregate damage, and not the SCC only.
To sum up, in order to compare the impact on SCC of dierent damage structures, we need to de ne a measure of aggregate damage. With this to-be-de ned measure, it becomes meaningful to say that, at an equal (aggregate) damage, the SCC is higher or lower when damage bears on growth rather than on production. The need of such a measure for controlled comparison is common in economics. For example, a comparison between constant discounting and hyperbolic discounting can be made by keeping the same degree of impatience, that is an equal discount of a constant stream of revenues (Strulik, 2015).
The rest of the paper is organized as follows. Section 2 presents the simple climate-economy model used and the three damage structures tested: conventional quadratic dam-age on production, linear damdam-age on economic growth and quadratic damdam-age on eco-nomic growth. Section 3 introduces our measure of the magnitude of damage, the damage strength, adapted from the Stern (2006) review. Section 4 compares how SCC varies across damage structures for an equal strength. We show that there is no unambiguous ranking of the structure, but that it depends mainly on the parameters of the social welfare function (the utility discount rate and the intertemporal elasticity of substitution). The comparison is then extended for the welfare gain of mitigation. Section 5 concludes with lessons learned and ways forward.
2 The climate-economy model
We use a very simple climate-economy model, based on the DICE model of Nordhaus (Nordhaus, 2008). The economic part is a copy from DICE. The climate module has been simpli ed from the DICE version, but with grounds in climate science (see the discussion below). To give an overview of this kind of model, it is a Solowian capital accumulation model with a climate feedback. There are two control variables: the saving rate, that controls the accumulation of capital, and the abatement rate, that controls the emissions and thus, through their accumulation, the temperature increase. Temperature increase feedbacks on production (or on TFP growth) through damage.
2.1 Economic module
At time t, production Qt is made from capital Kt and labor Lt through a Cobb-Douglas
function:
Qt= Ω(Tt)AtKtαL1−αt (1)
The total factor productivity (TFP) is At, and Ω(Tt) is the damage factor that reduces
current production due to temperature increase Tt; Ω is given by the following standard
function:
Ω(Tt) =
1 1 + π2Tt2
(2) On top of this conventional damage factor, in our model, climate change acts as a drag on TFP growth. The exogenous trend gt is reduced endogenously by current temperature
increase Tt:
At+1
At
= 1 + gAt(Tt) = 1 + gt− κ1Tt− κ2Tt2 (3)
When at least one of the κ is positive, this introduces a new type of damage: TFP growth rate is reduced and can even become negative when temperature increase is su ciently high. The two control variables are the gross saving rate st and the abatement (emission
control) rate µt. The capital stock is depreciated at rate δ and increases with investment:
Kt+1− Kt= −δKt+ Qt.st (4)
Emissions abatement costs a fraction Λt of the current production:
Λt= θ1(t)µθt2 (5)
where θ1(t) measures total costs and decreases exogenously, due to technical progress,
and θ2 is the xed exponent of the abatement rate µt in the abatement cost function,
representing that, at a given time, reducing a higher share of emissions becomes more and more expensive (θ2>1).
Total consumption is Ctand equilibrium between resources and uses imposes that:
Qt= QtΛt+ Qtst+ Ct (6)
2.2 Climate module
Emissions of greenhouse gases (here we just consider carbon dioxyde) are a by-product of production, partially oset by abatement µt.
Et= σt(1 − µt)Qt (7)
When there is no abatement µt = 0, emissions are proportional to production, with a
carbon content of production σt that decreases exogenously.
In most IAMs, emissions build up in the atmosphere. The carbon of the atmosphere is mixed between several carbon pools, which results in increasing concentrations. The
atmosphere temperature is set through interactions with ocean temperature and forcing from GHG. This machinery represents some of the feedbacks that rules climate change in the real world.
Here we keep only the essential link between emissions and temperature. We set tem-perature increase to be proportional to the cumulative emissions. Thus temtem-perature is simply given by:
Tt= β.(CE0+
X
0≤s<t
Es) (8)
where β is the proportionality factor and CE0 is the cumulative amount of emissions
between pre-industrial times and the starting time of the model, such that the term in the parenthesis is the cumulative emissions since pre-industrial times at time t.
Although this is a crude representation of the climate system, with several drawbacks4,
it has support in the climate community as the simplest rule of thumb to compute the reaction of global mean temperature to CO2 emissions accumulation. The ratio of global
warming to cumulative carbon dioxide emissions has been shown to be almost independent of time and of emissions pathway in simulations of the response to a range of emissions scenarios with carbon climate models, as well as in observations (Matthews et al., 2009; Gillett et al., 2013). The latest estimations yield an observationally constrained 5%95% range for the ratio of global warming to cumulative carbon dioxide emissions of 0.7 K/TtC to 2.0 K/TtC. Goodwin et al. (2015) also give a theoretical foundation to this near-linear dependence between warming and cumulative carbon emissions, with a theoretically de-rived equation of the dependence of global warming on cumulative carbon emissions over time. They show the proportionality results from compensating eects of oceanic uptake of heat and carbon. Their analysis identi es a surface warming response to cumulative carbon emissions of 1.5 ± 0.7 K/TtC.
We adopt this simple formula to ease computations but it is in no way an essential part of our argument.
2.3 Damage structures
In our model, climate damage has two main channels:
the damage factor (equation (2)), which reduces current production relative to the possible production with the same amount of capital, labor, and technology (TFP), but without climate change.
the impact of temperature increase on TFP growth (equation (3)), which reduces the technology (TFP) available in further period, compared to a case without climate change.
These direct channels feedback on the economic dynamics. These higher order ef-fects occur mainly through the accumulation of capital, most notably via reduced savings (Fankhauser and Tol, 2005). This has the important consequence that both damage on production and on TFP growth have long lasting eects and reduce the apparent growth rate of the economy. This is one of the reason why it is hard to distinguish between damage on production and damage on TFP growth in empirical studies, that observe only economic outcomes and not the channels per se.
4In particular, Krasting et al. (2014) have shown that this linear relationship is in uenced by prior
emissions pathways, and Leduc et al. (2015) have shown that it is no longer valid for high emissions pathways such as the RCP8.5.
To elucidate which kind of damage leads to the highest SCC, we consider three damage structures:
quadratic damage on production: π2 > 0 but κ1 = 0 and κ2 = 0. Note that in
this case there is no impact of climate change on TFP, so that the TFP evolves exogenously. This is the standard case commonly used in IAMs.
linear damage on growth: π2 = 0, κ1 > 0 and κ2 = 0. The damage factor Ω stays
equal to 1, so that there is no direct damage on production. The only channel for damage is the reduction in growth that occurs linearly in temperature. This is the case studied in Pindyck (2011).
quadratic damage on growth: π2 = 0, κ1 = 0 and κ2 > 0. There is again no
direct damage on production. Temperature increase impacts the TFP growth rate quadratically.
These three speci cations represent pure damage structures, in the sense that they do not mix impacts on production and on growth, and that the dependence in temperature is either linear or quadratic. Note that Dietz and Stern (2015) or Moyer et al. (2014) combine quadratic damage on production with quadratic damage on TFP growth, whereas Moore and Diaz (2015) include both linear damage on growth and linear damage on production. In reality, both types of damage, on production and on growth are possible, and probably co-exist. However, the aim of our contribution is not to study the most realistically calibrated damage, but instead to isolate the eect of changing the damage structure, while controlling for the damage strength. Previous studies that introduced damage on growth in an IAM did not study how the choices of the functional form and of the parameters values impacted the damage strength, nor compared dierent damage structures for the same strength.
3 Methodological discussion: measure of aggregate climate
damage and scenarios
To compare the impact of the damage structure, we need to measure the magnitude of climate damage in our model. This damage strength will be assessed against a bench-mark where there is no climate change; more precisely, it will be related to variations of intertemporal welfare. Thus the plan of this section: rst, we de ne the scenarios to be considered in our comparison; second, we explain how we measure dierences of welfare; third, we introduce the damage strength and related measures (mitigation gain, welfare loss).
3.1 Scenarios under scrutiny
We de ne three scenarios: the idealine5 (noted i), the baseline (b) and the optimal policy
(o). In each, we say precisely how the control variables µt and st are (endogenously)
computed.
The benchmark scenario, against which damage strength is to be measured, is a scenario in an ideal economy when there is no climate damage. As damage is the only feedback of climate change on this economy, this is equivalent to say that there is no climate change in the model. In this ideal world, there is obviously no need for mitigation (µt= 0), and
5What we call idealine is usually called baseline in the IAM community, in particular in the context of
the economy follows the optimal growth path where savings are optimized to maximise the intertemporal discounted sum of welfare. Thus the model reduces to the standard Ramsey-Solow model of capital accumulation. We call this scenario of optimal saving in a no-climate change economy the idealine, a contraction of ideal and baseline. The name reminds us that it is a pure hypothetical situation, and not a possible situation of the real world.
The baseline is the situation where climate damage impact the economy but no miti-gation is undertaken, for whatever reasons. Abatement µt is zero and gross saving rates
are those of the idealine. Contrary to the idealine, the baseline is a real possibility: it will be our world if no mitigation is ever implemented. This scenario could also be called laissez-faire scenario.
Finally, the third scenario to be considered is the optimal mitigation policy scenario: this is the scenario where the saving and abatement policy is chosen optimally, taking into account abatment cost and climate damage costs, to maximise intertemporal welfare.
The intertemporal welfare is given by the discounted utility ows weighted by popula-tion (so called total utilitarianism):
W (C) =X t≥0 Lt (1 + ρ)tu Ct Lt (9) where ρ is the utility discount rate, and u the utility function. The consumption stream Ct is endogenously computed from control variables thanks to equations (1) to (8).
This optimal scenario is the one traditionally studied in IAMs in cost-bene t analysis. The SCC is computed along the optimal scenario thanks to the formula:
SCC = − ∂EW u0C0
L0
(10)
where ∂EW represents the (negative) eect on welfare of an additional emission at date 0. 3.2 Measuring variations of welfare between scenarios
We are looking for a measure of welfare variation between scenarios.
Two scenarios, 1 and 2, have two corresponding consumption streams C1
t and Ct2, and
two intertemporal welfares W1 and W2. We will call the welfare variation of 2 compared to
1, denoted by w2|1, the proportional variation of the consumption stream C1 that provides the same welfare as in scenario 2, i.e. the quantity w2|1 such that:
W C1(1 + w2|1) = W (C2). (11)
This is the consumption loss, now and forever experienced in scenario 2 compared to reference scenario 1 (obviously w2|1< 0 if welfare in 2 is less than in 1).
This measure of welfare variation is known under dierent names. In climate change economics, it is the relative change in balanced growth equivalent, that Nicholas Stern (2006) introduced in his review, following his early studies on balanced growth equivalent as a measure of intertemporal welfare (Mirrlees and Stern, 1972). The welfare variation can be simply computed, when the utility function is iso-elastic with the formula (the elasticity of marginal social utility of (per-capita) consumption being noted θ):
w2|1=
W2
W1
1−θ1
When θ = 1, the formula is a little more complex, see Antho and Tol (2009) who have derived formulæ for several cases and computed growth equivalent variations with the FUND model.
In macro-economics, what we name welfare variation is called the welfare gain of a change from 1 to 2 (Lucas, 2003). It was famously applied by Lucas (1987, chap. III) on the welfare gain of eliminating the business cycle. The work of Pindyck (2011, 2012) uses the term willingness to pay (WTP) to avoid 2 while in 1. This is minus the welfare variation.
3.3 The strength of climate damage
Following Stern (2006), we de ne the damage strength as minus the welfare variation of the baseline compared to the idealine d = −wb|i. It is a measure of aggregate damage
whereas the SCC is a measure of marginal damage. The damage strength is the maximum proportion of idealine consumption (where there is no climate change) that we would be ready to sacri ce in order to avoid the baseline (where climate damage occurs but no mitigation is undertaken). Equivalently, we could also say that it is the proportion of idealine consumption lost due to the existence of unmitigated climate change. Because there is no bene t to climate change in our model, damage strength is necessarily positive. Comparing welfare in the idealine and the baseline is a good option to assess the damage strength, as it represents the maximum intertemporal damage that can aect the consumption path. It would make no sense to compare the idealine and the optimal policy. Indeed, in the latter, climate damage has been (partially) oset by mitigation. With our de nition of damage strength, the dierence of welfare arises just from climate damage and not from a combination of damage and mitigation policy.
Other welfare variations between the scenarios also have an economic meaning. The welfare variation between the optimal policy and the baseline can be named the mitigation gain g = wo|b. The mitigation gain is a measure of how much mitigation policy improves
welfare. It is the relevant measure to assess how eective mitigation policy can oset climate damage. Baseline and optimal policy are two speci cations of policy variables within the same model. Because the optimal policy is maximizing the welfare, it yields a higher welfare than the baseline policy, i.e. mitigation gains are positive.
The welfare variation of the optimal policy against the idealine is the residual welfare loss due to climate change as oset by mitigation policy l = −wo|i. The welfare loss
is measured relative to a constant idealine (this does not depend on the damage), so the welfare loss is an inverse measure of the absolute welfare in the optimal mitigation scenario. Due to the properties of welfare variation, damage strength, mitigation gain and welfare loss are related by the following relation:
1 − l = (1 + g)(1 − d). (13)
When there is almost no mitigation gain, climate damage on the baseline cannot be reduced and the welfare loss is equal to the damage strength. If, on the contrary, mitigation gain are important, climate damage can be totally oset, and welfare loss would be almost null. We will see how these quantities behave depending on the damage structure.
3.4 Damage structures, parameters and scenarios studied
We computed a number of runs of the model combining:
runs scenarios damage functionparameters social choiceparameters π2 κ1 κ2 ρ(%) θ
idealines (common
to all structures) idealine 0 0 0 1; 1.5; 2 1; 2; 3 quadratic damage
on production baselineoptimal pathway 00.02700.027 00 00 1; 1.5; 21; 1.5; 2 1; 2; 31; 2; 3 linear damage on
growth baselineoptimal pathway 00 00.0060-0.006 00 1; 1.5; 21; 1.5; 2 1; 2; 31; 2; 3 quadratic damage
on growth baselineoptimal pathway 00 00 00.00200.002 1; 1.5; 21; 1.5; 2 1; 2; 31; 2; 3
Table 1: Model runs computed in the study.
the damage structure (the point of impact (production or TFP growth) and their functional form),
the values of the parameters in the structure (π2, κ1 and κ2),
the values of the social choice parameters (ρ and θ).
The values of all other parameters are set to the values from DICE2007 (Nordhaus, 2008), and given in Appendix 2.
Table 1 summarizes the runs of the model computed for this study.
For the social choice parameters, we test a set of values, in the range of values recently used in IAM studies.
The disagreement about the value of the elasticity of marginal utility (θ) comes from the fact this parameter plays dierent roles in the discounted expected utility model: it simultaneously re ects preferences for intertemporal substitution, aversion to risk, and aversion to (spatial) inequality (Godard, 2009). Using estimates of the demand systems, Stern (1977) estimated its value around 2, with a range of roughly 0-10. More recently, in his Climate Change review (Stern, 2006), he advocated for a normative value equal to 1. In response to the Stern Review, Arrow (2007) suggest values around 2-3; and Dasgupta (2007) advocates for a value between 2 and 4. We chose to test the values 1, 2 and 3.
Although there exist some ethical grounds to justify very low values for the utility discount rate (Stern (2014a); Stern (2014b)), we chose to test a conservative range of values, recently used in IAM studies: 1%, 1.5% and 2%.
4 Results
This section presents the results of our investigation.
For the three damage structures, we varied the damage parameters (either π2, κ1 or
κ2, depending on the structure) and calculated the damage strength and the SCC in the
optimal scenario. For each damage structure, we thus obtained a curve that relates the SCC to the damage strength. With these curves, we can now compare whether a damage structure leads to a higher SCC for the same damage strength: this amounts to compare the position of a curve relative to the others.
For each combination of the welfare parameters, we plotted the SCC as a function of damage strength for the three structures. The dashed line with square marks represents quadratic damage on production. The solid line with circle marks represents linear dam-age on growth, whereas the solid line with square marks represents quadratic damdam-age on
growth. We have gathered all the graphs on gure 1: nine graphs with three curves on each. In gure 1, columns share the same pure time-preference rate ρ, whereas rows share the same elasticity of marginal utility θ. Both parameters increase from bottom left to top right. 0 2 4 6 8 10 0 100 200 300 ρ = 1.5% θ = 2 0 2 4 6 8 10 0 100 200 300 ρ = 1.5% θ = 3 0 2 4 6 8 10 0 100 200 300 ρ = 1.5% θ = 1 damage strength (%) 0 2 4 6 8 10 0 100 200 300 ρ = 1% θ = 2 SCC ($/t CO 2 ) 0 2 4 6 8 10 0 100 200 300 ρ = 2% θ = 2 0 2 4 6 8 10 0 100 200 300 ρ = 1% θ = 3 SCC ($/t CO 2 ) quadratic on production linear on TFP growth quadratic on TFP growth 0 2 4 6 8 10 0 100 200 300 ρ = 1% θ = 1 damage strength (%) SCC ($/t CO 2 ) 0 2 4 6 8 10 0 100 200 300 ρ = 2% θ = 3 0 2 4 6 8 10 0 100 200 300 ρ = 2% θ = 1 damage strength (%) Figure 1: SCC as a function of damage strength. Increasing ρ from left to right, increasing θfrom bottom to top.
The main result is that there is no unequivocal ranking of the dashed line with respects to the solid lines. This is su cient to conclude that having damage on growth does not necessarily lead to higher social cost of carbon than having damage on production, when damage strength is kept equal. This depends both on the welfare parameters and the precise functional forms for damage (linear or quadratic). This result puts into perspective previous works cited in the introduction. Their message was that SCC is higher when damage on TFP growth are taken into account. We show that this is not necessarily the case when the damage strength is controlled for. The explanation is that introducing
damage on TFP growth increases aggregate damage (see appendix 1). If we look more precisely at the curves, we can say that:
when damage is quadratic, SCC is higher when damage bears on production than when they bear on growth for the range of ρ and θ tested. Indeed the solid line with square marks is always below the dashed line.
Note that when θ = 1, there is almost no dierence.
the comparison between damage on production and linear damage on TFP growth depends on the welfare parameters. So whether having damage on growth induces higher SCC depends strongly on the social welfare function with which outcomes are evaluated. We can see that when θ is high (= 3), the SCC is higher when damage is on production. But when θ is low (= 1), the SCC is higher when damage is on growth. For intermediate θ (= 2), the ranking depends on the rate of pure time-preference: the dashed line is below the solid line with circle marks when ρ is low and above when ρ is high. With linear damage on growth, putting more weight to long run, i.e. lowering ρ or θ, is su cient to have higher SCC.
These results make an intuitive sense. Indeed, for the same damage strength, damage on TFP growth is more spread out across time than damage on production (see gure 4). To have the same intertemporal aggregate damage, damage on production at a given date must be higher in the short to medium run, whereas damage on growth is mainly concentrated in the long and very long run. Because the SCC is the discounted sum of marginal impacts, putting more weight on the long run, i.e. lowering ρ or θ, will tend to produce a higher SCC with damage on growth, relative to the SCC with damage on production. Indeed, at high θ, the SCC with damage on production is well above the SCC with damage on growth, whereas at low θ, the SCC with damage on production is below the SCC with linear damage on growth and slightly above, but very close to, the SCC with quadratic damage on growth. With linear damage on growth, damage is higher in the medium run than with quadratic damage on growth, which leads to higher SCC. The curves are also sensitive to ρ but much less than to θ.
To sum up, there is no straightforward relationship between the damage structure and the SCC. Having damage on growth rather than on production does not increase the SCC: the eect depends on the welfare parameters.
This situation for the SCC, with no unequivocal relationship, can be contrasted with the comparison of the welfare loss l between the optimal policy and the idealine, as de ned in the fourth paragraph of section 3.3. By de nition, the welfare loss, that compares welfare between the optimal policy and the idealine, will be lower than the damage strength, that compares welfare between the no-mitigation policy (baseline) and the idealine6. Therefore,
the relation between welfare loss and damage strength will stay below the bisectrix of the rst quadrant.
The welfare loss between optimal policy and the idealine as a function of damage strength is depicted in gure 2, with the same convention as in gure 1. Here we have a straightforward ranking between the three damage structures under scrutiny, for the range of parameters explored. For a given damage strength, the nal welfare, after mitigation, will be higher with quadratic damage on growth and lower with linear damage on growth. Quadratic damage on production is a median situation. Alternatively, we can say that
mitigation reduces damage more when damage are quadratic than linear, and more when they bear on TFP growth than on production.
For a given damage strength, damage will not be distributed with the same temporal pro le for dierent damage structures (see gure 4). With quadratic functional form, damage is quite low at the beginning, but increase on the medium term. Thus it is mainly concentrated in the long-run. With linear functional form, damage is substantial from the very beginning. Damage is thus more uniformly spread across time. Now, a mitigation policy has almost no eect in the short run, this means that it can mostly oset damage in the long-run. As the long-run damage takes a greater share of the overall level of damage with a quadratic functional form, welfare loss is less important in this case than with a linear functional form. Regarding the dierence between quadratic damage on production and on growth, the eect is the same, as damage on growth mainly occurs in the very long term, they can be more easily oset by a mitigation policy.
As shown in gure 3, for the mitigation gain between the optimal policy and the baseline, there is also a straightforward ranking between the three damage structures under scrutiny, for the range of parameters explored. For a given damage strength, the welfare gain due to mitigation is higher with quadratic damage on growth and lower with linear damage on growth. Quadratic damage on production is a median situation, and almost coincides with the case of quadratic damage on growth when θ = 1. The intuition for this result is the same as for welfare loss: mitigation action can better oset damage that occur in the long term.
5 Conclusion
In this article, we compared, in a simple climate-economy model, the eect of three damage structures on the value of the SCC. We found that the ranking of SCC between a model with damage on production and a model with damage on TFP growth is not unequivocal. It depends on welfare parameters such as the utility discount rate or the elasticity of marginal social utility of consumption. Quadratic damage on growth does give a lower SCC than quadratic damage on production, for the range of welfare parameters we considered; however, the comparison of quadratic damage on production and linear damage on growth depends on the welfare parameters. When the pure time-preference rate and the elasticity of marginal utility are low, the SCC is higher with damage on growth. On the contrary, when pure time-preference rate and the elasticity of marginal utility are high, damage on production gives a higher SCC. Based on recent econometric studies, several authors asserted that climate damage is higher than previous evaluations (see Tol (2009, 2014) for a review of these previous evaluations), and imply higher SCC because they aect TFP and not only production. We do not challenge this point. However, it can be inferred from our results that the higher SCC values come in fact from two eects: one linked to higher aggregate damage than previously evaluated, the other linked to the location of damage on TFP.
To compare damage structures, we have followed Stern (2006) and de ned a measure of aggregate damage based on welfare variation, that we called damage strength. It allows disentangling the role of the magnitude of damage (the damage strength) from the role of the representation of damage, what we have called a damage structure. The rst one is the magnitude of climate damage: damage can be low or high. The second one is the location of damage: on production, on TFP growth, as analyzed in this article, but also maybe on capital, or on working hours, on population, etc. These are two dierent questions because a high magnitude of damage can be represented within a damage structure on
0 2 4 6 8 10 0 2 4 6 8 ρ = 1.5% θ = 2 0 2 4 6 8 10 0 2 4 6 8 ρ = 1.5% θ = 3 0 2 4 6 8 10 0 2 4 6 8 ρ = 1.5% θ = 1 damage strength (%) 0 2 4 6 8 10 0 2 4 6 8 ρ = 1% θ = 2 w elfare loss (%) 0 2 4 6 8 10 0 2 4 6 8 ρ = 2% θ = 2 0 2 4 6 8 10 0 2 4 6 8 ρ = 1% θ = 3 w elfare loss (%) quadratic on production linear on TFP growth quadratic on TFP growth 0 2 4 6 8 10 0 2 4 6 8 ρ = 1% θ = 1 damage strength (%) w elfare loss (%) 0 2 4 6 8 10 0 2 4 6 8 ρ = 2% θ = 3 0 2 4 6 8 10 0 2 4 6 8 ρ = 2% θ = 1 damage strength (%)
Figure 2: Welfare loss l between the optimal policy and the idealine, as a function of damage strength. Increasing ρ from left to right, increasing θ from bottom to top.
production only, whereas a low magnitude of damage can just as well be represented within a damage structure on TFP growth. Measuring aggregate damage at only one reference point (for example 2.5◦C) as Kopp et al. (2012) or only in a given timeframe as Pottier
et al. (2015), is not appropriate because it does not account for the whole trajectory of damage, especially when dierent damage structures are involved. The damage strength we use has the advantage of accounting for this whole trajectory.
However, the measure of damage strength has one main limitation: it is not observable and the link with empirical data is thus not direct. One would hope that empirical studies would be able to identify simultaneously the magnitude of damage and its structure. In the term of our model, empirical studies would then be able to give values to the triplet of
0 2 4 6 8 10 0 2 4 6 8 10 ρ = 1.5% θ = 2 0 2 4 6 8 10 0 2 4 6 8 10 ρ = 1.5% θ = 3 0 2 4 6 8 10 0 2 4 6 8 10 ρ = 1.5% θ = 1 damage strength (%) 0 2 4 6 8 10 0 2 4 6 8 10 ρ = 1% θ = 2 mitigation ga in (%) 0 2 4 6 8 10 0 2 4 6 8 10 ρ = 2% θ = 2 0 2 4 6 8 10 0 2 4 6 8 10 ρ = 1% θ = 3 mitigation gains (%) quadratic on production linear on TFP growth quadratic on TFP growth 0 2 4 6 8 10 0 2 4 6 8 10 ρ = 1% θ = 1 damage strength (%) mitigation gain (%) 0 2 4 6 8 10 0 2 4 6 8 10 ρ = 2% θ = 3 0 2 4 6 8 10 0 2 4 6 8 10 ρ = 2% θ = 1 damage strength (%)
Figure 3: Mitigation gain g between the optimal policy and the baseline, as a function of damage strength. Increasing ρ from left to right, increasing θ from bottom to top.
π2, κ1, κ2 (and potentially values for parameters representing damage located on channels
not represented in our model but that are likely to exist in reality, e.g. damage on capital, on working hours, on population, etc.) that represents the true nature of climate damage in the real world. However the relationship between the representation of damage in an IAM and empirical data is more complex.
First, the identi cation of econometric eects is not an easy task, depends on many assumptions and remains debated. Indeed, econometric studies rely on economic output (GDP) and economic growth (ratio of GDP between two years). What is observed is a time series of economic (GDP) growth rates. From this point of observation, all damages struc-tures considered in this article boil down to a reduction of yearly growth rate. Inferring the damage structure from observations of economic growth data is thus not a straightforward
2015 2050 2100 2150 2200 2250 −40 −30 −20 −10 0 date consumption loss (%) quadratic on production linear on TFP growth quadratic on TFP growth
Figure 4: Consumption loss between baseline and idealine for three damage structures. Damage strength is 3%, ρ = 1.5% and θ = 2.
task, and relies necessarily on disputable choices and assumptions, especially regarding the functional forms chosen and the identi cation strategies. Econometric studies used variations of annual temperature across space or, more preferably, across time. An instan-taneous eect of a shock of annual temperature is denominated a production impact of climate change and a persistent eect is denominated a (GDP) growth impact of climate change. Strictly speaking, it does not identify parameters of the climate damage structure, but parameters of the response to weather changes. Whether the response of societies to changing weather conditions is a good proxy for their response to climate change is still controversial (see for example the diering positions expressed in Auhammer et al. (2013), Dell et al. (2014) and Hsiang (2016), see also Schneider (1997) for an early discussion of this debate). Therefore, the empirical identi cation of climate damage, whether on production or on GDP growth, is not yet rmly established in the econometric literature.
Second, it is currently not possible to relate directly the coe cients of econometric regression to the parameters of a damage structure in an IAM. Indeed, there is potentially not a unique set of parameters of the IAM to reproduce empirical observation, due to the plausible co-existence of a number of channels of damage (on output, on growth, on capital, on working hours, on population...) that are not all estimated jointly in econometric studies. Therefore, parameters are under-determined. Furthermore, the representation of feedbacks in an IAM structure (e.g. through saving rates, or with TFP increasing endogenously with the capital stock) changes the ultimate eect of damage on actual GDP growth, such that parameters of damage functions have to be changed from the econometric coe cients to reproduce observations. Moore and Diaz (2015) is a rst attempt to calibrate parameters of damage functions in an IAM on econometric results (from Dell et al. (2012)), which illustrates the fact that parameters are not directly equal to econometric coe cients. Despite valuable insights provided by empirical studies of economic impacts, we are still a long way from a comprehensive identi cation of the damage parameters to be used in the IAMs. To realistically calibrate the representation of damage in IAMs, further methodological and empirical investigations are critical. In the mean time, dierent choices for the representation of damage in an IAM will co-exist. This is the reason why we need
to address the kind of question asked in this paper: what is the impact on the SCC of dierent damage structures, given a magnitude of damage. When discussing climate damage, authors should distinguish between the impact on the SCC and the impact on the damage strength, between the impact on the marginal damage and the impact on aggregate damage. To do so, a good practice would be to systematically report the SCC along with the damage strength or another measure of the magnitude of damage. If the damage strength cannot be used to directly make the link between models and empirical observation, it can be reported in modelling studies to disentangle how much of the results is due to the magnitude of damage, and how much is due to the very nature of the alternative representation tested.
Appendix 1
Moyer et al. (2014) and Dietz and Stern (2015) rely on the same representation of the impact of climate change on productivity growth. They start from damage bearing only on production, with a damage function Dt that depends quadratically on the temperature
Tt: Dt= 1 − Ω(Tt). Net production (that is including climate damage) Qt is reduced by a
factor 1 − Dtfrom Yt, the production that would have occured given factors on production
(technology, capital, labor), had climate change not existed: Qt= Yt(1 − Dt)
From this standard case, the current damage Dt are then allocated between damage
on production and damage on TFP growth, a procedure that originates, to our knowledge, from Kopp et al. (2012). The TFP growth rate is reduced by f.Dt, when the production,
instead of being reduced by 1 − Dt is reduced by (1 − Dt)/(1 − f Dt), where f is the
share of damage that impact growth. Speaking of an allocation of damage conveys the impression that there is the same amount of damage.
However the damage apply on very dierent quantities. Damage on TFP growth have a ( rst-order) cumulative eect on the whole output path, whereas damage on production does not7. Far from keeping aggregate damage constant across the scenarios studied, the
allocation introduces more damage when f increases. Thus, it is not surprising that these studies nd that the SCC increases when more damage are allocated to damage on growth. To illustrate this, we consider two possible ways to measure aggregate damage: the real production loss and the damage strength used in the main text.
Figure 5 plots the real production loss at 3◦C(the percentage of production loss between
the baseline and the idealine when the temperature increase reaches 3◦C) against the
theoretical loss (the value of Dtwhen the temperature increase is 3◦C), for several shares
of damage on growth: 0, 2.5, 5, 10, 15 and 20 (percents). Because temperature increase feedbacks on the economic dynamic and thus on emissions, the date at which 3◦Cis reached
changes slightly when the theoretical loss and the share vary.
When damage bear only on production, real losses are already higher than the theoret-ical damage from the damage function (equal theorettheoret-ical and real damage is represented by the dotted line in gure 5). This is due to feedbacks through reduced accumulation of capital. But the dierence between real and theoretical losses remains quite small. When the share of damage on TFP growth increases, the wedge between the real loss and the theoretical loss increases rapidly. This is of course no surprise, because the damage that reduces long-term growth rates has long-lasting eects. However it has the consequence that damage is much higher at 3◦Cthan what is commonly assumed. This means that the
value of reduction of output at 3◦Cis not kept constant.
7As explained in the main text, damage on production have a second-order cumulative eect through
0 5 10 15 20 25 0 20 40 60 80 0 0 2.5 2.5 5 5 10 10 15 20
theoretical production loss (%)
real
pro
duction
loss
(%)
Figure 5: Real production losses as a function of theoretical production losses, for dierent shares of damage on growth f increasing from 0 to 20 percents.
0 5 10 15 20 25 0 20 40 60 80 100 0 0 2.5 2.5 5 5 10 10 15 15 20
theoretical production loss (%)
damage
stren
gth
(%)
Figure 6: Damage strength as a function of theoretical production losses, for dierent shares of damage on growth f increasing from 0 to 20 percents.
Figure 6 plots the damage strength against the theoretical production loss. We can see that theoretical loss and damage strength goes in the same direction. Therefore, when the share of damage on TFP growth is kept constant, theoretical loss can be used as a proxy for damage strength. However, when the share of damage on TFP growth is changed, the same theoretical loss can lead to highly dierent damage strengths.
Appendix 2 (online)
The model is resolved over a 600 years time horizon, by 5 years time steps. It is calibrated in 2005. Table 2 gives the parameters values.
In addition, four variables follow exogenous trends, as de ned below: gt= g0e−χt (exogenous trend of TFP growth)
parameters values units names
A0 0.02722 - Initial level of TFP
L0 6514 million Initial world population
K0 137 trillion $ Initial value of world capital
α 0.3 - Capital elasticity in production function δ 10% - Depreciation rate of capital per year θ2 2.8 - Exponent of abatement cost function
β 2 K/TtC Parameter linking cumulative emissions and temperature increase
Table 2: Values of model parameters.
Lt= L0e−γt+ L∞(1 − e−γt) (population)
σt= σ0e−gσte
−dσ t 1+e−speakdpeak
1+espeak(t−dpeak) (exogenous evolution in carbon content of production)
θ1(t) = σt
pbackstop
θ2
rbackstop−1+e−gbackstopt
rbackstop (exogenous decrease of abatement costs)
Table 3 gives the values of the parameters used in these four exogenous trends.
parameters values units names
g0 0.92% - Initial growth rate of TFP per year
χ 0.1% - Rate of decline of exogenous trend of TFP growth per year
γ 3.5% - Initial growth rate of population per year L∞ 8600 million Asymptotic population
σ0 0.00013418 tC/$ Initial carbon content of production
gσ 0.73% - Initial decrease rate of carbon content of pro-duction per year
dσ 0.3% - Rate of decline of decarbonization per year
speak 0.03 per year Speed of fast decline of carbon content
dpeak 150 year Date of fast decline of carbon content
pbackstop 1170 $/tC Cost of backstop
rbackstop 2 - Ratio initial to nal backstop cost
gbackstop 0.5% - Initial decline in cost of backstop per year
Table 3: Values of parameters de ning exogenous trends.
References
Anthoff, David and Richard SJ Tol (2013) The uncertainty about the social cost of carbon: A decomposition analysis using fund, Climatic Change 117 (3), pp. 515530. Anthoff, David and Richard S. J. Tol (2009, July) The Impact of Climate Change
on the Balanced Growth Equivalent: An Application of FUND, Environmental and Resource Economics 43 (3), pp. 351367.
Arrow, Kenneth J. (2007, January) Global Climate Change: A Challenge to Policy, The Economists' Voice 4 (3).
Auffhammer, M., S. M. Hsiang, W. Schlenker, and A. Sobel (2013, July) Using Weather Data and Climate Model Output in Economic Analyses of Climate Change, Review of Environmental Economics and Policy 7 (2), pp. 181198.
Colacito, Riccardo, Bridget Hoffmann, and Toan Phan (2014, December) Temper-atures and Growth: A Panel Analysis of the U.S., SSRN Scholarly Paper ID 2546456, Social Science Research Network, Rochester, NY.
Committee on Assessing Approaches to Updating the Social Cost of Car-bon, Board on Environmental Change and Society, Division of Behavioral and Social Sciences and Education, and National Academies of Sciences, Engineering, and Medicine (2017) Valuing Climate Changes: Updating Estima-tion of the Social Cost of Carbon Dioxide, Washington, D.C.: NaEstima-tional Academies Press, DOI: 10.17226/24651.
Dasgupta, Partha (2007) The Stern Review's economics of climate change, National institute economic review, pp. 47.
Dell, Melissa, Benjamin F. Jones, and Benjamin A. Olken (2009) Temperature and Income: Reconciling New Cross-Sectional and Panel Estimates, American Economic Review 99 (2), pp. 198204.
Dell, Melissa, Benjamin F. Jones, and Benjamin A. Olken (2012, July) Temper-ature Shocks and Economic Growth: Evidence from the Last Half Century, American Economic Journal: Macroeconomics 4 (3), pp. 6695.
Dell, Melissa, Benjamin F. Jones, and Benjamin A. Olken (2014) What Do We Learn from the Weather? The New ClimateEconomy Literature, Journal of Economic Literature 52 (3), pp. 740798.
Deryugina, Tatyana and Solomon M. Hsiang (2014, December) Does the Environ-ment Still Matter? Daily Temperature and Income in the United States, Working Paper 20750, National Bureau of Economic Research.
Dietz, Simon and Nicholas Stern (2015, March) Endogenous Growth, Convexity of Damage and Climate Risk: How Nordhaus' Framework Supports Deep Cuts in Carbon Emissions, The Economic Journal 125 (583), pp. 574620.
Fankhauser, Samuel and Richard S.J. Tol (2005, January) On climate change and economic growth, Resource and Energy Economics 27 (1), pp. 117.
Gillett, Nathan P., Vivek K. Arora, Damon Matthews, and Myles R. Allen (2013, March) Constraining the Ratio of Global Warming to Cumulative CO2 Emissions Using CMIP5 Simulations*, Journal of Climate 26 (18), pp. 68446858.
Godard, Olivier (2009) Time discounting and long-run issues: the controversy raised by the Stern Review of the economics of climate change, OPEC energy review 33 (1), pp. 122.
Goodwin, Philip, Richard G. Williams, and Andy Ridgwell (2015, January) Sensi-tivity of climate to cumulative carbon emissions due to compensation of ocean heat and carbon uptake, Nature Geoscience 8 (1), pp. 2934.
Greenstone, Michael, Elizabeth Kopits, and Ann Wolverton (2013, January) Developing a Social Cost of Carbon for US Regulatory Analysis: A Methodology and Interpretation, Review of Environmental Economics and Policy 7 (1), pp. 2346. Hope, Chris (2013) Critical issues for the calculation of the social cost of CO2:
why the estimates from PAGE09 are higher than those from PAGE2002, Climatic Change 117 (3), pp. 531543.
Hsiang, Solomon (2016, October) Climate Econometrics, Annual Review of Resource Economics 8 (1), pp. 4375.
Hsiang, Solomon M. and Amir S. Jina (2014, July) The Causal Eect of Environmental Catastrophe on Long-Run Economic Growth: Evidence From 6,700 Cyclones, Working Paper 20352, National Bureau of Economic Research.
Interagency Working Group on Social Cost of Carbon (2010) Social Cost of Carbon for Regulatory Impact Analysis under Executive Order 12866., Technical report, United States Government.
Interagency Working Group on Social Cost of Carbon (2015) Technical Update of the Social Cost of Carbon for Regulatory Impact Analysis Under Executive Order 12866., Technical report, United States Government.
Kopp, Robert, Richard Tol, and Stephanie Waldhoff (Eds.) (2012) The Social Cost of Carbon, http://www.economics-ejournal.org/special-areas/special-issues/the-social-cost-of-carbon. Special Issue of Economics: The Open-Access, Open-Assessment E-Journal.
Kopp, Robert E., Alexander Golub, Nathaniel O. Keohane, and Chikara Onda (2012) The In uence of the Speci cation of Climate Change Damages on the Social Cost of Carbon, Economics: The Open-Access, Open-Assessment E-Journal.
Krasting, J. P., J. P. Dunne, E. Shevliakova, and R. J. Stouffer (2014, April) Trajectory sensitivity of the transient climate response to cumulative carbon emissions, Geophysical Research Letters 41 (7), pp. 25202527.
Leduc, Martin, H. Damon Matthews, and Ramón de Elía (2015, December) Quan-tifying the Limits of a Linear Temperature Response to Cumulative CO 2 Emissions,
Journal of Climate 28 (24), pp. 99559968.
Lucas, Jr, Robert E. (1987) Models of business cycles, Oxford, Royaume-Uni: Basic Blackwell.
Lucas, Jr, Robert E. (2003, March) Macroeconomic Priorities, The American Eco-nomic Review 93 (1), pp. 114.
Matthews, H. Damon, Nathan P. Gillett, Peter A. Stott, and Kirsten Zickfeld (2009) The proportionality of global warming to cumulative carbon emissions, Na-ture 459 (7248), pp. 829832.
Mirrlees, James A. and Nicholas H. Stern (1972, April) Fairly good plans, Journal of Economic Theory 4 (2), pp. 268288.
Moore, Frances C. and Delavane B. Diaz (2015, February) Temperature impacts on economic growth warrant stringent mitigation policy, Nature Climate Change 5 (2), pp. 127131.
Moyer, Elisabeth J., Mark D. Woolley, Nathan J. Matteson, Michael J. Glotter, and David A. Weisbach (2014, June) Climate Impacts on Economic Growth as Drivers of Uncertainty in the Social Cost of Carbon, The Journal of Legal Studies 43 (2), pp. 401425.
Nordhaus, W. D. (1992) An optimal transition path for controlling greenhouse gases, Science 258 (5086), pp. 13151319.
Nordhaus, William D. (1994) Managing the Global Commons: The Economics of Climate Change, Cambridge (Mass.): MIT Press.
Nordhaus, William D. (2008) A Question of Balance, London: Yale University Press. Nordhaus, William D. (2010) Economic aspects of global warming in a post-Copenhagen environment, Proceedings of the National Academy of Sciences 107 (26), pp. 1172111726.
Pindyck, Robert S. (2011) Modeling the Impact of Warming in Climate Change Economics, in G. Libecap and R. Steckel (Eds.), The Economics of Climate Change. Chicago: University of Chicago Press.
Pindyck, Robert S. (2012, May) Uncertain outcomes and climate change policy, Jour-nal of Environmental Economics and Management 63 (3), pp. 289303.
Pizer, William, Matthew Adler, Joseph Aldy, David Anthoff, Maureen Crop-per, Kenneth Gillingham, Michael Greenstone, Brian Murray, Richard Newell, Richard Richels, Arden Rowell, Stephanie Waldhoff, and Jonathan Wiener (2014, May) Using and improving the social cost of carbon, Science 346 (6214), pp. 11891190.
Pottier, Antonin, Etienne Espagne, Baptiste Perrissin-Fabert, and Patrice Dumas (2015) The comparative impact of Integrated Assessment Models' structures on optimal mitigation policies, Environmental Modelling & Assessment 20 (5), pp. 453 473.
Revesz, Richard L., Peter H. Howard, Kenneth Arrow, Lawrence H. Goulder, Robert E. Kopp, Michael A. Livermore, Michael Oppenheimer, and Thomas Sterner (2014) Global warming: Improve economic models of climate change, Nature 508 (7495), pp. 173.
Riahi, Keywan, Detlef P. van Vuuren, Elmar Kriegler, Jae Edmonds, Brian C. O'Neill, Shinichiro Fujimori, Nico Bauer, Katherine Calvin, Rob Dellink, Oliver Fricko, Wolfgang Lutz, Alexander Popp, Jesus Crespo Cuaresma, Samir KC, Mar-ian Leimbach, Leiwen JMar-iang, Tom Kram, Shilpa Rao, Johannes Emmerling, Kristie Ebi, Tomoko Hasegawa, Petr Havlik, Florian Humpenöder, Lara Aleluia Da Silva, Steve Smith, Elke Stehfest, Valentina Bosetti, Jiyong Eom, David Gernaat, Toshi-hiko Masui, Joeri Rogelj, Jessica Strefler, Laurent Drouet, Volker Krey, Gunnar Luderer, Mathijs Harmsen, Kiyoshi Takahashi, Lavinia Baumstark, Jonathan C. Doelman, Mikiko Kainuma, Zbigniew Klimont, Giacomo Marangoni, Hermann
Lotze-Campen, Michael Obersteiner, Andrzej Tabeau, and Massimo Tavoni (2017) The Shared Socioeconomic Pathways and their energy, land use, and green-house gas emissions implications: An overview, Global Environmental Change.
Schneider, Stephen H. (1997, December) Integrated assessment modeling of global climate change: Transparent rational tool for policy making or opaque screen hiding value-laden assumptions?, Environmental Modeling & Assessment 2 (4), pp. 229249. Stern, Nicholas (1977) Welfare weights and the elasticity of the marginal valuation of
income, in M. Artis and A. Nobay (eds), Studies in Modern economic Analysis, Oxford: Blackwell.
Stern, Nicholas (2014a) Ethics, equity and the economics of climate change paper 1: Science and philosophy, Economics and Philosophy 30 (03), pp. 397444.
Stern, Nicholas (2014b) Ethics, equity and the economics of climate change paper 2: Economics and politics, Economics and Philosophy 30 (03), pp. 445501.
Stern, Nicholas H. (2006) Stern Review: The Economics of Climate Change, Vol-ume 30, London: HM treasury.
Strulik, Holger (2015, January) Hyperbolic discounting and endogenous growth, Economics Letters 126, pp. 131134.
Tol, Richard SJ (2009) The economic eects of climate change, The Journal of Economic Perspectives, pp. 2951.
Tol, Richard SJ (2014) Correction and Update: The Economic Eects of Climate Change, The Journal of Economic Perspectives 28 (2), pp. 221225.
